Research On Arcing Fault Protection Methods Under Nonlinear Loads

With the development of the smart grid,there are more and more kinds of household electrical equipment,especially nonlinear electrical equipment.The change in the types of load puts forward higher requirements for the safety of electricity.Arcing fault is an important cause of the electric fire,including two types of series and parallel.Due to the limitation of the line loads,the current of the series arc fault is usually less than the normal working current of the load,and common circuit protection devices such as circuit breakers and residual current protectors cannot effectively protect them,which constitutes a hidden peril of the electric fire.The current waveform of the nonlinear loads is similar to that of the arcing fault,and the current of the arcing branch may be coupled with that of the normal load branch,which further increases the difficulty in the detection of arcing fault and leads to error identification.The thesis carries out research on the method of series arc fault protection under the nonlinear loads,and solves the problem of the protection of arcing fault under known and unknown loads.(1)Refer to the requirements of GB / T 31143,an arcing fault experimental platform is built to conduct and analyze the arcing fault under typical linear loads,non-linear loads,and their combined conditions.(2)An arcing fault detection method based on time-frequency characteristics of current is studied.Db 5 wavelet is selected for the pretreatment of the arcing current.The difference of the cosine similarity of the approximation coefficient of the adjacent periodic current is used as the characteristic quantity of the low-frequency,and the characteristic quantity in the high-frequency domain is the wavelet energy in 3125 Hz-6250 Hz frequency bandwidth.On this basis,an arcing fault recognition algorithm based on threshold comparison is proposed.The results show that the proposed method can accurately identify the arcing fault under a single load and combined loads.There is no error identification during the startup of the nonlinear loads.(3)Based on the research of the arcing fault protection method based on current timefrequency domain characteristics,further study of the identification method is based on deep learning.Compare and analyze common networks that are used in deep learning,select the AlexNet for the identification of arcing fault and improve the size of its convolution kernel to reduce parameters of the original model.Use the training strategy with adaptive learning rate adjustment to improve the convergence speed.Use the stochastic gradient descent method to optimize the weight update process and reduce the training time.Use the arc fault data to train and verify the improved AlexNet.The results show that the accuracy of the arcing fault identification method under known and unknown loads reaches 98.5% and97.5% respectively.At the same time,the parameters of the proposed model are reduced by28% compared with those of the traditional AlexNet.Both of the proposed methods achieve the identification and protection of arcing fault and have certain advantages and disadvantages respectively.The traditional method based on the time-frequency domain characteristics of arcing current has relatively small algorithm complexity,which is beneficial to the implementation of embedded hardware.However,this method requires artificial determination of the protection action threshold and has the limitation of the identification of the arcing fault under unknown loads.The improved AlexNet method does not need to determine the threshold manually,has stronger load adaptability.It can identify the arcing fault under unknown loads,but the algorithm is relatively complex and requires relatively high hardware performance.